76 D. H. HYERS [February A NOTE ON LINEAR TOPOLOGICAL SPACES* D. H. HYERS A space T is called a linear topological space if (1) T forms a linear f space under operations x+y and ax, where x,yeT and a is a real number, (2) T is a Hausdorff topological space,J (3) the fundamental operations x+y and ax are continuous with respect to the Hausdorff topology. The study § of such spaces was begun by A. Kolmogoroff (cf. [4]. Kolmogoroff's definition of a linear topological space is equivalent to that just given). Kolmogoroff calls a set 5 c T bounded, if for any sequence xveS and any real sequence av converging to 0 we have lim,,.^ avxv = 0, where 6 is the zero element of T. He then shows that a linear topological space T reduces to a linear normed space|| if and only if there exists in T an open set which is both convex^ and bounded. In this note, the characterization of other types of spaces among the class of linear topological spaces is studied. Spaces which are locally bounded, that is spaces containing a bounded open set, are found to be "pseudo-normed" on the one hand, and metrizable on the other, but not in general normed. Fréchet «paces, or spaces of type (F), are characterized. The main result of the paper is that a linear topological space T is finite dimensional, and hence linearly homeomorphic to a finite dimensional euclidean space, if and only if T contains a compact, open set. This of course is a generaliza­ tion to linear topological spaces of the well known theorem of F. Riesz for the space of continuous functions. We first give some needed properties of bounded sets. The follow­ ing notations will be used throughout. We denote by aS the set of all ax with xeS; by x + 5, the set of all elements x+y where y ranges over S; by S1 + S2, the set of all x+y with xeSi, yeSz. THEOREM 1. A set S of a linear topological space is bounded if and only if, given any neighborhood U of the origin, there is an integer v such that \a\ <l/v implies aSc U. * Presented to the Society, April 3, 1937. t Cf., for example, [l ], p. 26. Numbers in brackets refer to the bibliography at the end of the paper. t Cf. [2], pp. 228-229, axioms (A), (B), (C), (5); or [3], pp. 43 and 67. § Important instances of a linear topological space were studied by J. von Neu­ mann several years before Kolmogoroff's paper was published. Cf. Mathematische Annalen, vol. 102 (1930), pp. 370-427. See also [5]. || Cf. [l],p.53. 1T A set S is convex if and only if x,yeS and 0 <a<l imply ax-\-(l — a)yeS. License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use 1938] LINEAR TOPOLOGICAL SPACES 77 PROOF. The condition of the theorem is sufficient, for suppose that it is satisfied. Then if a\—>0, x\eS, we see, for any chosen neighborhood U of the origin 0, that a\X\eU for all X for which | a\\ < 1/v. That is, \im\^a\X\ = d. The condition is necessary. For, assume that S is bounded, but that the condition of the theorem does not hold. Then for some U and any integer v>0, there exists xveS, and a real av with |a„| <l/v such that avxvlU. Therefore a„—>0 with 1/V, but avxv does not converge to 6. That is, S is not bounded. This contradiction proves the necessity of the condition. This theorem provides us with an alternative definition of bounded- ness. For other equivalent definitions see [5] and [ó]. See also [7]. THEOREM 2. A compact set of a linear topological space is bounded. PROOF. Let 5 c T be compact, so that every infinite subset of 5 has a limit point in T. Assume, contrary to the theorem, that S is not bounded. Let V be any chosen neighborhood of the origin 6. By denying the definition of boundedness, we are led to a sequence av—*0 and a sequence xveS, with avxveU, for p = l, 2, • • • . But, since 5 is compact, there is a limit point p of the infinite sequence xv. By the continuity of the function ax at a = 0, x = p, there exists a 8>0 and a neighborhood V of p such that aVc U for \a\ <S. Since p is a limit point of the sequence x„, there is an infinite subsequence x„K such that xVKeV for all K. For sufficiently large K we have |ce„J <ô, and therefore aVlixVlieaVKV c U, contrary to the fact that avxvlU. This contradiction proves the theorem. DEFINITION. A linear space E will be said to be pseudo-normed if corresponding to each xeE there is a real number \x\ with the properties : (i) |#( ^0; \x\ =0 implies x — di (ii) I ax I = |a| \x\ (where |a| is the absolute value of a), (iii) if \x\—»0, \y\—>0, then |x+;y|—K). A linear topological space T will be called pseudo-norniable if it can be pseudo-normed in such a way that the topology according to the pseudo-norm is equivalent to the original Hausdorff topology. THEOREM 3. Every pseudo-normed linear space is a locally bounded linear topological space. Conversely, if a linear topological space con­ tains a bounded open set, then the space is pseudo-normable. PROOF: Let E be a pseudo-normed space and for any chosen a consider the "sphere" \y — x\ <a. Denote by U(x\ a) the set of points z of this sphere for each of which there exists a positive ô such that License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use 78 D. H. HYERS [February I w — zI <5 implies |w—#| <a. It is easy to verify that E is a Haus­ dorff space with neighborhoods U(x; a), and that the operations of vector addition and scalar multiplication are continuous, so that E is a linear topological space. Property (iii) of the pseudo-norm is used in proving both the Hausdorff separation axiom and the continuity of addition. Local boundedness now follows from property (ii). On the other hand, let T be a locally bounded linear topological space, and suppose that U is a bounded open set. Without loss of generality we may suppose that U contains the origin, and that U=( — 1)U. Then the sets all, a^O, are all open, and clearly form a complete neighborhood system of the origin. Put \x\ =g.l.b. \a\, xealf, and it follows without difficulty that the postulates for a pseudo-normed space are satisfied. Next, for any integer v>0 con­ sider the set \x\ <1/V. By definition of \x\, this is the set of all x such that xealf, \a\ <l/v, and hence* the set \x\ <l/v is open. Now by Theorem 1 and the fact that U is bounded, for any neighborhood F of the origin there is an integer v = v(V) such that \a\ <l/v implies aUc V. Hence the sphere \x\ <1/V, being the union of the sets all for which | a \ <l/*>, is contained in V, and the spheres \x\ < 1/V form a complete neighborhood system of the origin. Thus the topology according to the pseudo-norm is equivalent to the original Hausdorff topology. COROLLARY 1. A locally bounded linear topological space satisfies the first countability axiom. COROLLARY 2. If T is locally bounded, then a set S cT is bounded if and only if there is a JJL>0 such that xeS implies \x\ </x. THEOREM 4. A necessary and sufficient condition that a linear topological space be finite dimensional is that it contain a non-empty set which is both open and compact. PROOF. The necessity is immediate, since every finite dimen­ sional linear topological space (a linear space with a "finite basis" (xi, • • • , xv) such that every element is uniquely expressible in the form ^2\=IOÙ\X\) is linearly homeomorphic to a finite dimensional euclidean space.f To prove the sufficiency, let U be a compact open set containing the origin. By Theorem 2, V is bounded, and hence T is pseudo-normable by Theorem 3. If 5 is any bounded set of T, * [7], Theorem 1.2. Special use is made of the continuity of ax in proving this theorem. t For the proof see [8]. License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use m*] LINEAR TOPOLOGICAL SPACES 79 then for some a we have S call (Theorem 1), so that S is compact, being a subset of the compact set aU. Also, from Corollary 1, every limit point is the limit of a convergent sequence. Hence every bounded sequence has a convergent subsequence, where the con­ vergence is taken according to the pseudo-norm. From here on, with the aid of Corollary 2, the proof follows that of a similar result for the linear normed space of continuous functions due to F. Riesz,* and will be omitted. COROLLARY 3. A linear topological space T is linearly homeomorphic to a finite dimensional euclidean space if and only if T is locally compact. A theorem on metrizability of "Hausdorff groups" has been proved by G. Birkhoff (see [9]), which states that a Hausdorff group is metrizable if and only if the first countability axiom is satisfied.